为什么我的Spark Streaming应用程序在微型批次的UI中显示0条记录？

Question

我编写了一个自定义InputDStream，并在compute函数中创建了RDD，方法是从HBase中读取内容，然后进行一些映射，例如：

  class DataStream(ssc: StreamingContext, seedOpt: Option[Long] = None) extends InputDStream[Array[Byte]](ssc) {
    private var random: Random = _
    private var rdd: RDD[(Int, Array[Byte])] = _
    private var rddCreationTime: Time = _

    private val config = Config("mytable", "colfamilty", 3, 10)
    private val numImportanceLevels = 10
    private val importanceLowerFraction = 0.1
    private val importanceUpperFraction = 0.8
    private val fractions = linearImportanceFractions(
      importanceLowerFraction, importanceUpperFraction, numImportanceLevels
    )

    override def start(): Unit = {
      seedOpt match {
        case Some(seed) => random = new Random(seed)
        case None => random = new Random()
      }
    }

    override def stop(): Unit = {

    }

    private def needRecreateRdd(currentTime: Time): Boolean = {
      if (rdd == null) true
      else {
        val duration: Duration = currentTime - rddCreationTime
        duration > Seconds(60)
      }
    }

    private def getRdd(currentTime: Time) = {
      if (needRecreateRdd(currentTime)) {
        rddCreationTime = currentTime
        val endTime = ZonedDateTime()
        val beginTime = endTime.minusDays(30)
        rdd = config.createTrainingRddWithImportance(ssc.sparkContext, beginTime, endTime, numImportanceLevels).cache()
      }
      rdd
    }

    override def compute(validTime: Time): Option[RDD[Array[Byte]]] = {
      //println(s"----> $validTime")

      val samples = getRdd(validTime)
        .sampleByKey(withReplacement = false, fractions, random.nextLong())
        .map({case (importance, tfRecord) => tfRecord})

      Some(samples)
    }
  }

它工作正常，我可以从Python运行程序中获取字节数据（我正在使用TensorflowOnSpark）。但是，如果我检查Spark UI的“流式传输”选项卡，则据说我所有的迷你批处理都有0条记录：

那是为什么？